Re: [PATCH] Add rubygems updater.

[Ben Woodcroft]

On 03/01/16 06:54, Ludovic Courtès wrote:
> Ben Woodcroft <address@hidden> skribis:
>
> > On 02/01/16 04:17, Ludovic Courtès wrote:
> > > Ben Woodcroft <address@hidden> skribis:
> > >
> > > > On 01/01/16 19:28, Pjotr Prins wrote:
> > > > > On Fri, Jan 01, 2016 at 06:27:21PM +1000, Ben Woodcroft wrote:
> > > > > > It seems there's 30 packages to be updated, out of the 107 in
> > > > > > ruby.scm. Going through each of these individually seems a little
> > > > > > tedious, can we do them in bulk somehow or do they have to be
> > > > > > committed individually? Building and testing all packages that
> > > > > > require these packages would be a start - is there any way to list
> > > > > > all dependent packages?
> > > > > >
> > > > > > gnu/packages/ruby.scm:2807:13: ruby-cutest would be upgraded from
> > > > > > 1.2.2 to 1.2.3
> > > > > > gnu/packages/ruby.scm:333:13: ruby-rspec-mocks would be upgraded
> > > > > > from 3.2.1 to 3.4.0
> > > > > (etc)
> > > > >
> > > > > I don't think it is a good idea to automatically update
> > > > > packages. Reason being that packages should be updated by someone who
> > > > > is actively using that new version. Automated tests are one thing,
> > > > > real user feedback another. Not to mention that many gems don't have
> > > > > tests ;).
> > > > I think we should update the package definitions so that more have
> > > > tests, and failing that import the library so we know it can at least
> > > > be loaded, like this:
> > > >
> > > > +     `(#:phases
> > > > +       (modify-phases %standard-phases
> > > > +         (replace 'check
> > > > +           (lambda _
> > > > +             (zero? (system* "ruby" "-Ilib" "-r" "ansi")))))))
> > > The only case where this would make a difference is for leaf packages,
> > > no?  In all the other cases, building dependent packages will ensure
> > > that the package at hand works as expected.
> > Sure, but even in the case where they aren't leaf packages at least
> > the build error gets thrown when building the package at
> > fault. There's also the important difference that it makes the
> > packager feel less bad about the disappointing lack of tests or the
> > necessity of disabling them because of circular dependencies.
> Right.  The only downside I can think of is if packagers have to copy
> the above 4 lines in each and every package.  Can you think of a way
> that would avoid that?
I have only been adding these in cases where testing is impossible, but 
we could make it a wider policy.

We could bake it into the build system, by adding an optional argument 
#:import so that you could do

    (build-system ruby-build-system)
    (arguments
     `(#:import "ansi"
       #:tests? #f)) ; tests require circular dependencies

Probably in that case makes sense to have a new phase 'check-import so 
that more complex cases can be handled, rather than replacing 'check. 
There's no way to run this phase with the native-inputs disappeared is 
there so it more closely mirrors a user's experience?

We could even default this to the expected name of the library guessed 
from the name of the package when #:import is not given. However, this 
would unfortunately break packages that have been written outside of 
Guix, so I imagine you don't feel this is a good idea.

WDYT?

Thanks,
ben